Figure 03 demonstrates autonomous ladder climbing in video
US based Figure has shared a video that it says shows its Figure 03 humanoid robot climbing a ladder fully autonomously, according to Interesting Engineering. The footage was posted on X by founder Brett Adcock and shows the robot ascending a ladder without visible human assistance.
The claim is technically relevant because ladder climbing stresses more than basic walking. A humanoid has to coordinate arms and legs, maintain balance while shifting load between contact points, and perceive where rungs and feet are in relation to its body. Figure has not disclosed the control stack used in the clip, and the demonstration has not been independently verified.
A demanding mobility test with limited disclosure
Adcock posted the video with the statement, “F.03 can now climb a ladder, fully autonomous.” The source article says the clip drew attention from AI researchers, with some describing it as “good progress” in humanoid robotics.
For operators and robotics teams, the useful point is narrower than the social media reaction. Ladder climbing is a high coordination task that can expose weaknesses in whole body control, perception latency, contact planning, and recovery behavior. A single video does not show repeatability, fall handling, speed, payload, power draw, or whether the ladder and environment were constrained for the test.
Adcock also used a separate post to argue that “wheeled robots are an utter dead end,” restating his view that legged humanoids are better suited to human built environments including factories, warehouses, offices, and homes. That is a familiar argument in the humanoid sector, but the ladder clip by itself does not settle the tradeoff between legged systems and simpler mobile platforms in industrial settings.
Figure ties the demo to Helix updates
Figure describes Figure 03 as a general purpose humanoid for everyday use and says its Helix AI model is intended to let the robot navigate unpredictable home environments. The company has also recently described an upgrade to Helix System 0, or S0, that combines visual perception with whole body motion control.
According to the source article, the earlier system relied on proprioception, meaning the robot’s internal awareness of joint positions, body motion, and balance. Figure says the updated S0 model adds real time visual data from onboard stereo cameras, processes RGB images, and builds a three dimensional representation of the surrounding environment. The company says that should support more precise foot placement, improved balance, and smoother movement across terrain such as stairs, ladders, and uneven surfaces.
Figure also says the model was trained end to end with reinforcement learning in simulation across randomized terrains and environmental conditions, and that the learned behaviors transfer to physical robots without additional calibration or fine tuning. Those claims are central to the ladder demo, but they remain company supplied. Figure has not published enough detail in this report to evaluate training coverage, failure cases, or how much of the behavior was constrained by the setup.
The company recently said it increased Figure 03 production from one unit per day to one per hour and had delivered more than 350 robots. If accurate, that production claim gives the autonomy work a larger fleet context, since more hardware in circulation can accelerate data collection and iteration. The next useful evidence would be repeated ladder runs, varied ladder geometry, and full disclosure of autonomy conditions.
Source: interestingengineering.com
